Automatic speech recognition system using seed templates

Abstract

An automatic speech recognition system has a multi-mode training capability using a set of previously stored templates of a limited number of predetermined seed words to train the templates for a vocabulary of words. The training speech samples each includes a vocabulary word juxtaposed with a seed word. An averager module maintains an active average template for each of the word units of the training speech samples including the seed word units, and the active average templates are used to continuously update the seed template set as they are used in the training speech samples. The preferred training procedure employs training phrases each having a vocabulary word embedded between two seed words, and two seed template sets are used in succession, the first being composed of single-digit words, and the second composed of carrier words.

I claim: 
     
       1. A training subsystem for an automatic speech recognition system having a training capability for training a vocabulary of words to be recognized by said system, comprising: a seed template set for maintaining templates of template parameters for a limited set of seed words which are preselected to be short, commonly-used, and easily recognized words;   a training input for providing a training speech sample for training each vocabulary word to be recognized by the automatic speech recognition system, wherein each training speech sample consists of a spoken phrase of a vocabulary word juxtaposed with at least one seed word included in said seed template set;   an extractor for extracting template parameters for each of the words of a training speech sample provided by said training input, wherein said extractor is enabled to extract the template parameters for the vocabulary word of the training speech sample by using the template maintained in said seed template set for the at least one seed word of the training speech sample;   a training control module for controlling said extractor to provide the extracted template parameters for each vocabulary word of the respective training speech samples and for generating corresponding vocabulary word templates;   a dictionary storage for storing the templates for the respective vocabulary words as extracted by said extractor and generated under control of said training control module; and   said training control module being operative for controlling said extractor to provide the extracted template parameters for the at least one seed word of the training speech sample and for updating the corresponding seed word template of said seed template set so that the updated seed word template can be used for subsequent training speech samples.   
     
     
       2. A training subsystem for an automatic speech recognition system according to claim 1, wherein each training speech sample provided by said training input is composed of a vocabulary word bracketed on each side by a seed word of said seed template set. 
     
     
       3. A training subsystem for an automatic speech recognition system according to claim 1, wherein said training control module is operative for training a vocabulary of words for any one of a plurality of applications, speakers, and recognition modes for which the system is used. 
     
     
       4. A training subsystem for an automatic speech recognition system according to claim 1, wherein said training control module includes means for using two seed template sets in succession, the first being composed of seed words which are single-digit words, and the second composed of seed words which are short carrier words having alternate types of phoneme suffixes and prefixes. 
     
     
       5. A training subsystem for an automatic speech recognition system according to claim 4, wherein said training control module includes means for commanding a training procedure of training the templates for the single-digit words as a first seed template set, training the templates for the carrier words as a second seed template set using the first seed template set of digit words as the maintained seed template set, then training the vocabulary words using the second seed template set of carrier words as the maintained seed template set. 
     
     
       6. A training subsystem for an automatic speech recognition system according to claim 5, wherein said training control module includes means for further commanding the training procedure of training the vocabulary words using phrases composed of the words of a particular application bracketed by the seed words, then verifying recognition of the vocabulary words using phrases composed of the vocabulary words in the syntax of the particular application. 
     
     
       7. A training subsystem for an automatic speech recognition system according to claim 1, further comprising an averager module operative in conjunction with said extractor and said training control module for averaging the template parameters for the words of a training speech sample repeated a plurality of times and for maintaining active average templates corresponding to such words of the training speech sample, wherein said averager module is used to generate averaged templates for the respective vocabulary words and to continuously update active average templates of the seed template set as the respective seed words are used in successive speech training samples. 
     
     
       8. A method for training a vocabulary of words to be recognized by an automatic speech recognition system, comprising the steps of: maintaining a seed template set of templates of template parameters for a limited set of seed words which are preselected to be short, commonly-used, and easily recognized words;   providing a training speech sample for training each vocabulary word to be recognized by the automatic speech recognition system, wherein each training speech sample consists of a spoken phrase of a vocabulary word juxtaposed with at least one seed word included in said seed template set;   extracting template parameters for each of the words of a training speech sample, wherein extracting the template parameters for the vocabulary word of the training speech sample is enabled by using the template maintained in said seed template set for the at least one seed word of the training speech sample;   using the extracted template parameters for each vocabulary word of the respective training speech samples to generate corresponding vocabulary word templates;   storing the templates for the respective vocabulary words in a dictionary storage to be used by the automatic speech recognition system for recognizing vocabulary words; and   further using the extracted template parameters for the at least one seed word of the respective training speech samples to update the seed word templates of said seed template set so that the updated seed word templates can be used for subsequent training speech samples.   
     
     
       9. A method for training a vocabulary in an automatic speech recognition system according to claim 8, further comprising the step of averaging the template parameters for the words of a training speech sample repeated a plurality of times and maintaining active average templates corresponding to such words of the training speech sample, wherein averaged templates are generated for the respective vocabulary words and active average templates are maintained for the seed template set in order to continuously update the seed template set as the respective seed words are used in successive speech training samples. 
     
     
       10. A method for training a vocabulary in an automatic speech recognition system according to claim 8, wherein each training speech sample is composed of a vocabulary word bracketed on each side by a seed word of the seed template set. 
     
     
       11. A method for training a vocabulary in an automatic speech recognition system according to claim 8, wherein the training method is used to train the vocabulary words for any one of a plurality of applications, speakers, and recognition modes for which the system is used. 
     
     
       12. A method for training a vocabulary in an automatic speech recognition system according to claim 8, wherein said extracting step includes using two seed template sets in succession, the first being composed of seed words which are single-digit words, and the second composed of seed words which are short carrier words having alternate types of phoneme suffixes and prefixes. 
     
     
       13. A method for training a vocabulary in an automatic speech recognition system according to claim 12, wherein said training method includes commanding a training procedure of training the templates for the single-digit words as a first seed template set, training the templates for the carrier words as a second seed template set using the first seed template set of digit words as the maintained seed template set, then training the vocabulary words using the second seed template set of carrier words as the maintained seed template set. 
     
     
       14. A method for training a vocabulary in an automatic speech recognition system according to claim 13, wherein said training method includes further commanding the training procedure of training the vocabulary words using phrases composed of the words of a particular application bracketed by the seed words, then verifying recognition of the vocabulary words using phrases composed of the vocabulary words in the syntax of the particular application.